In a database system, data is stored in a form of page. When a page is concurrently accessed, a lock mechanism needs to be used to ensure consistency. A read lock blocks a write operation on the page. A write lock blocks read and write operations on the page. When blocking occurs, it is referred to as contention on the page. The contention on the page causes a data access latency, which is also referred to as a sleep state, and therefore, total performance of the database is severely affected. Key factors that cause contention on a page are: there is data of more than one line on a same page; a write operation is frequently performed on at least one line; read and write operations are frequently performed on at least another line.
Identifying a hot page and taking a corresponding measure, which is generally to split content on the hot page into a plurality of pages, can reduce contention on the page, and improve the total performance of the database system.
Currently, only contention on a page, that is, an absolute value of a quantity of times of access to the page, is considered in the prior art, a real hot page cannot be identified, and only a page with a largest quantity of times of access can be identified.